ChatGPT

[sighyoung]

This has been the first academic year in which a lot of students seemed to turn in essays based partly on material partially drafted on ChatGPT or similar systems.

Right now, it's a little difficult to separate out writing problems caused by high school preparation (this is still the cohort that was affected by the pandemic in high school), but I am seeing some patterns that are a bit more pronounced that in the past.

1) Introductions to paragraphs are now universally terrible. Vague opening paragraphs have long been a problem (as well as papers that state the topic rather than state the argument), but increasingly, students can't zero-in on their main arguments later in the essay. I suspect that's partly because they fed the essay prompt into ChatGPT, got what they got, and couldn't really restate, development, or elaborate on ideas that weren't really theirs to beginning with.

Many good introductions can hint at the structure of the essay, but most students couldn't do that at all.

2) In connection with introductions, titles are terrible. It's a common problem, but it's worsened because students don't seem to know the main argument of their essays.

3) Topic sentences have almost disappeared from writing. Most students this semester wrote a transitional sentence at the beginning of the paragraph but couldn't state the main argument of the paragraph.

4) Irrelevant background information about the author. My students are supposed to write a short five-page essay about a short story or poem, but my God, they go on and on about Emily Dickinson's personal life in Amherst, Massachusetts, which is not what the assignment is about. In the past, one or two students would do this because they didn't read carefully or know what to say, but it's an epidemic now.

5) Formulaic or paint-by-the-numbers arguments that aren't clearly related to the main idea of the essay. So students will write in the first paragraph "to show this [whatever "this" is], Author X uses characters, imagery, and symbolism," which is a statement of facts rather than an argument about the kinds of imagery, say, or how that imagery develops and shifts meaning over the course of the poem. Then paragraph two will say "Characters are really important in this play," which isn't an argument, but a fact. But there is little detail to show that the student read the story--he/she/they might have, but there's no detail here.

6) Structural balance is off in many papers. Introductions, for instance, are either way too short or far too long for the length of the essay. Increasingly, there's a lost sense of proportion.

7) Instead of lots of examples, the student will put in one gigantic quotation which is supposed to prove everything within a paragraph. (This is likely do to taking an inherited structure and adding a specific example to it, rather than developing the argument organically.) Again, in the past, maybe two students in a class would do this, but now several do it.

8) The worst of the papers have invented quotations and events in the story because AI generated crap for them. This is especially likely when the poem or story is obscure. I'm less likely to get crap about a Faulkner short story like "A Rose for Emily," but if it's a little-known poem by an obscure Harlem Renaissance writer, there is a non-negligible chance that AI will spit out something made up.

9) While I'm unlikely to catch students plagiarizing, class averages are declining. I will say that if a paper contains made-up quotations for a ten-line poem, I have a really strong case for academic dishonesty, but, in general, grades in some classes were way down this semester. Usually, I have mostly A's and B's, with a couple of C's and maybe three F's because people never turned in work.

This semester, in my online class, there were only 6 A's or higher in a class of 24. The quality is just clearly down.

[cardinalkarp]

Very interesting Sigh, apparently the internet is now making people dumb and lazy.

When I went back to school essay writing was such a large part of the curriculum, and at this point it seems to have been outsourced to AI by many students. Is there a way to “test” more and have students write papers less?

For the lazy college student AI basically made their life way easier (granted the quality doesn’t seem to be there), but at some point I was wondering what it’s going to do to higher education (hell, even high school education).

[Leroy]

I was dumb and lazy way before internets.

[BottenFieldofDreams]

I’ve been playing around with the update. They’re unironically saying it’s like Her, like that’s good. And maybe the well was poisoned, but it does sound a lot like ScarJo. I hope she sues, because I’d like to see these guys collect some Ls before they destroy us. I could see it being useful, like this morning if it could have told me if I’m allowed to add extra bags to the yard trimming pickup bin all the time or just some of the time. I don’t think I’ll use it much.

Anyway, time to start applying for some plumbing apprenticeship programs.


I also watched the futurist Amy Webb’s talk at SXSW on YouTube. Kind of interesting, though it seemed like she jumped to very random scenarios for AI fallout. One thing she talked about that was interesting/scary was the way face computers will track pupil dilations to collect extremely personal data. It’s like the way Doritos are designed to keep you eating. These folks are going to win a lot of control over people. A lot more, I guess.

Plus side is I worry about Trump v Biden less. I think it’s a less consequential battle than ANYONE v AI. Or less optimistically, who to root for (though I doubt it matters much) within AI. My kids are going to have [expletive] -ing saddles on their back.


ETA: only the demo ‘pick your voice’ part sounded like ScarJo. Oddly the actual use voice was pretty different.

[BottenFieldofDreams]

So I’m not the only one that thinks so. Johansson and her lawyers agree it sounds like her.

Apparently OpenAI was trying to get her to partner with them. She said no and they did it anyway. Great omen for those of us that want to have jobs.

They’re going to [expletive] us. It’s just too bad we’ve already seen Uber and AirBnB and Amazon and all the social media companies do whatever they want and get away with it. Because otherwise maybe there’d be some hope that they aren’t just going to do whatever they want to do and pretend it’s not for the money.

[ghostrunner]

Yeah. Idk, maybe I'm a luddite now. There's no internal governor on these people and I hate their whole mindset.

[AdmiralKird]

So I’m not the only one that thinks so. Johansson and her lawyers agree it sounds like her.

Looks like Johansson filed a lawsuit. OpenAI has paused the ScarJo-like voice, "Sky" update while they investigate how they created it.

On the one hand, she doesn't really have a case if OpenAI managed to create a voice by using another actor and a whole amalgamation of vocal tone algorithms, but if they did something to have the AI specifically watch "Her" and dial it in to sound like Scarlett, then, eh, guess that would come out in discovery. Sounds like OpenAI wants to at least make sure they didn't step on any existing laws.

Also, obligatory:

[Arthur Dent]

So I’m not the only one that thinks so. Johansson and her lawyers agree it sounds like her.

Apparently OpenAI was trying to get her to partner with them. She said no and they did it anyway. Great omen for those of us that want to have jobs.

They’re going to [expletive] us. It’s just too bad we’ve already seen Uber and AirBnB and Amazon and all the social media companies do whatever they want and get away with it. Because otherwise maybe there’d be some hope that they aren’t just going to do whatever they want to do and pretend it’s not for the money.

It sure seems like we are all just going to stand by helplessly while we let some of the world's biggest psychopaths control the development of extremely dangerous technology for personal enrichment.

[BottenFieldofDreams]

As a Type 1 Diabetic I feel like I may benefit from AI at a higher impact than most people. It’s easy for me to dream up the ways it could help me live better and longer. Even essentially as someone that produces insulin.

At the same time, robots have the best pancreas, which is no pancreas at all. They never miss a day to go pee in a cup to measure ketones. This thing is talking in a natural sounding voice at damn near conversational speeds. It will get better and soon be less fallible than me. I always do worry too much about everything. But this is very troubling. In a way the extinction level event talk only provides smoke cover for no effing jobs.

[ghostrunner]

Tried to ask Chat GPT to give a full list of position players who have pitched this season and it couldn't do it accurately at all. Figured it would be a good use for GPT if it's accurate, preventing me from having to browse through a bunch of stats.

It gave a list of 34 players, and then when I came back with two innaccuracies, it gave me a reduced list, also inaccurate.

I specifically asked when Garrett Hampson pitched, because he was listed and I could find no games where he pitched.

It came back with:

Garrett Hampson pitched for the Kansas City Royals on June 14, 2024. He was brought in as a position player to pitch in a blowout game against the Cincinnati Reds​ (MLB.com)​​ (MLB.com)​.

The Royals haven't played the Reds this season. They played the Dodgers on that day, and Hampson didn't even play. The two links it provided at the end were links to stories about his doubles in two different games against the Guardians and the Royals on 6/30 and 7/4.

It's funny that it gets that specific for something that didn't happen.